KR20030046532A - Decomposition and synthesis method of english phonetic symbols - Google Patents

Abstract

PURPOSE: A decomposition and synthesis method of English phonetic symbols is provided to decrease the size of voice data, and to keep intonation of original language by decomposing phonetic symbols and sorting corresponding to intonation. CONSTITUTION: A decomposition and synthesis method of English phonetic symbols includes phoneme decomposition, continuant processing, intonation sorting and synthesis of speech. An English word is converted into phoneme series to decompose phonemes(1), and the continuant is inserted to the continuant position in phoneme series during continuant processing(2). Phonemes are sorted according to intonation(3), and voice waveform data of phoneme corresponding to phoneme series are synthesized to speech(4). Phonetic symbols are decomposed into basic phoneme elements, and voice is synthesized according to decomposed phoneme series. The phoneme decomposition step includes preprocessing, prefix consonant processing, processing consonant between vowels, and suffix consonant processing.

Description

DECOMPOSITION AND SYNTHESIS METHOD OF ENGLISH PHONETIC SYMBOLS}

TECHNICAL FIELD The present invention relates to the decomposition and synthesis of English phonetic symbols, and more particularly, to a method applied to a voice database and a speech synthesis system.

Verbal conversation is the verbal exchange of information between humans or between a human and a machine through a particular device. Speech communication includes conventional telephone conversations, different information services over the telephone, and various forms of language systems.

Linguistic systems usually use language databases or linguistic analysis and synthesis techniques. The language database is stored as a predetermined voice recorded by a real person. For example, a quiz database prepares a question and answer system. The user enters a predetermined question number via the keyboard, and the system finds and transmits the corresponding language in the language database. The advantage is that the real person records and the voice is clear. The drawbacks are the need for a large amount of storage and the limitation of the output from the system.

The linguistic analysis and synthesis technique is a linguistic technique according to a predetermined method for analyzing language signals and providing essential characteristic parameters synthesized into languages according to some language generation models. Typical applications are linguistic analysis and synthetic telephones. The method may also be used in an automated voice response system provided for an automated voice information service. The information service includes automatic retrieval in flight schedules, train schedules, market price quotes, and the like. People use the keyboard to query and get responses from the response system. Language analysis and synthesis systems have the advantage of greatly compressing data and facilitating communication encryption. However, the disadvantage is that stress, intonation, and syllabilization in language synthesis are unnatural and ambiguous.

Therefore, how to maintain the natural voice properties in language analysis and synthesis is the most important problem to be solved.

The present invention is K.K. It provides a method of decomposing phonetic symbols and classifying them according to intonation. The method decomposes known phonetic symbols into basic phonetic units of various parts, and then combines the basic phonetic units to form a voice according to a syllable combining method.

The present invention includes phoneme decomposition, continuous processing, intonation classification and language synthesis. The method converts the phonetic symbols of a word into a phoneme series and places the continuation sounds at consecutive sound positions in the phoneme series. After classifying intonations of phoneme strings, language waveforms are synthesized from the phoneme strings.

It is an object of the present invention to reduce the data storage space of the original language and to maintain the intonation of the original language.

Table 1 shows K.K. Conversion table between phonetic symbols and English characters.

1 is a flowchart illustrating a method of decomposing and synthesizing English phonetic symbols.

2 is a flowchart showing the phoneme decomposition step in detail.

3 is a flowchart detailing a processing step.

4 is a flowchart showing the prefix consonant process in detail.

5 is a flowchart showing in detail consonant processing between vowels.

6 is a flowchart showing the suffix consonant process in detail.

7 is a flowchart showing the continuous sound processing in detail.

8 is a flowchart showing the intonation classification in detail.

9 is a flowchart showing the initial classification in detail.

10 is a flowchart detailing the double stress processing.

11 is a flowchart detailing advanced classification.

The present invention is English K.K. It is related to phonetic symbols. Since the phonetic symbols and the English characters do not exactly match, there is a difficulty in word processing in the present specification. In the following description, English characters representing phonetic symbols are used. See FIG. 1 instead of the conversion table.

Referring to FIG. 1, the present invention is a method for decomposing and synthesizing English phonetic symbols, which includes phoneme decomposition (1), continuous sound processing (2), intonation classification (3), and language synthesis (4). .

The phoneme decomposition step (1) converts phonetic symbols of English words into phoneme strings. The continuous sound processing step 2 adds the continuous sound symbol to the continuous sound position of the phoneme string. The intonation classification step 3 classifies intonation of the phoneme string. The language synthesis step 4 synthesizes language waveforms based on phoneme strings.

Referring to FIG. 2, the phoneme decomposition step 1 includes a preprocessing 11, a prefix consonant processing 12, a vowel consonant processing 13, and a suffix consonant 14. Processing steps.

Referring to FIG. 3, the preprocessing step 11 converts a phonetic symbol into a phonetic symbol code and combines, inserts, or replaces the phonetic symbol code to calculate the number of consonants.

The pretreatment step 11 includes the following steps:

(a) Converting phonetic symbols to phonetic symbol codes (step 111).

(b) double consonant processing (step 112): [tsh], [dsg], [pr], [kr], [gr], [pl], [bl], [kl], [gl] double consonants, such as [tw], [hw], [sw], [dw], [kw] and [gw], are treated as a single consonant that cannot be separated.

(c) Suffix Consonant Processing (Step 113): The consonants at the ends of [ts] and [dz] and the like are treated as a single consonant which cannot be separated.

(d) Continuous Consonant Processing (Step 114): The continuous consonants are formed of consonants other than [r] following [m], [n] or [l]. If a consecutive consonant is at the end of a word, or after a first accent, a second accent, or a consonant, then a consonant other than [r] precedes the vowel [eh] and [m], [n], or [l] first Insert [eh] between consecutive consonants.

(e) Voiceless consonant processing (step 115): If [s] is not a suffix consonant, then unvoiced consonants [p], [t] or [k] after [s] are voiced consonants [b], [ d] or [g], respectively.

(f) Count the number of consonants (step 116).

Consonants can be classified as prefix consonants, vowel consonants and suffix consonants.

4, the prefix consonant processing step 12 determines whether a phoneme-separating position is located before the prefix consonant. If located, add diaeresis to the phoneme separation location. In the prefix consonant processing step 12, the phoneme separation position is located between the second prefix consonant and the final prefix consonant.

The prefix consonant processing step 12 includes the following steps.

(a) Checking that the number of prefix consonants is at least 1 (step 121). If not, the process ends.

(b) checking for a first prefix consonant (step 122). If no, continue to step (d).

(c) Select the next prefix consonant (step 123).

(d) Add diacritics before the prefix consonant (step 124).

(e) Check for final prefix consonants (step 125). If so, the process ends, and otherwise, the flow returns to step (c).

For example, the pronunciation symbol [fri] of "free" is treated as [f-ri], the pronunciation symbol [sprihng] of "spring" is [s-prihng], and the pronunciation symbol [strit] of "street" [s-drit].

Referring to Fig. 5, the intervowel consonant processing step 13 determines whether the phoneme separation position is located before the intervowel consonant. If located, diacritics are added. In step 13, the phoneme separation location is located before the consonant between the final vowels and before the consonant between vowels that do not belong to the set consisting of [m], [n], [l], [r] and [ng].

The consonant processing step 13 between vowels includes the following steps:

(a) It is checked whether the number of consonants between vowels is 1 or more (step 131). If not, check that the number is one. If 1, continue to step (e). If not (if 0), the process ends.

(b) It checks whether the consonant between vowels belongs to the set consisting of [m], [n], [l], [r] and [ng] (step 132). If not, a diacritical sign is added before the consonant between the vowels (step 133).

(c) Find consonants between the next vowel (step 134).

(d) Check for consonants between the final vowels (step 135). If no, continue to step (b).

(e) Inserting diacritics before consonants (step 136).

Referring to Fig. 6, the suffix consonant processing step determines whether the phoneme separation position is located before the suffix consonant. If located, add diacritics to the phoneme separation location. In step 14, the phoneme separation location is located before the final suffix consonant and before the suffix consonant that does not belong to the set consisting of [m], [n], [l], [r] and [ng].

Suffix consonant processing step 14 includes the following steps:

(a) Check the number of suffix consonants (step 141). If the number is zero, the process ends.

(b) Check if the suffix consonants belong to the set consisting of [m], [n], [l], [r] and [ng] (step 142). If not, a diacritic is added before the suffix (step 143).

(c) Find the next suffix consonant (step 144).

(d) Check for final suffix consonants (step 145). If there is, the process ends, and if not, it continues to step (b).

For example, the pronunciation symbol ['aebehkehs] of "abacus" becomes [ae-beh-keh-s] after processing. The pronunciation symbol [eh, brivih'eshehn] of "abbreviation" becomes [eh-bri-vihe-ahehn] after processing. The pronunciation symbol [aih, dentehfeh'keshehn] of "identification" becomes [aih-dean-teh-feh-ke-shehn] after processing. The pronunciation symbol [firhst] of "first" becomes [firh-s-t] after the processing.

Referring to FIG. 7, the continuous sound processing step 2 performs continuous sound processing on a consonant which is a follow-upconsonant of the previous phoneme and a leading consonant of the next phoneme. In the absence of the first accent or the second accent, as well as in the absence of consonants [m], [n], [l], [r] and [ng], etc. between the two vowels, such consonants are the continuation of the preceding and following phonemes Well. The consonant may be a consonant of the previous phoneme and a preceding consonant of the next phoneme.

The continuous tone processing step 2 includes the following steps:

(a) Check whether the number of phonemes is 1 or more (step 21). If not, the process ends.

(b) Select a second phoneme (step 22).

(c) It is checked whether the first phonetic symbol belongs to the set consisting of [m], [n], [l], [r] and [ng] (step 23). If not, continue to step (f).

(d) Check whether the final phonetic symbol of the previous phoneme belongs to the vowel and the set consisting of [l] and [r] (step 24). If no, continue to step (h).

(e) A continuous phonetic sign and a continuous phonetic consonant are inserted between the previous phoneme and the current phoneme (step 25).

(f) Select the next phoneme (step 26).

(g) Check for final phoneme (step 27). If it is correct, the process is terminated, otherwise, the process continues to step (c).

(h) Check whether the last symbol of the previous phoneme belongs to the set consisting of [m], [n]] and [ng] (step 28). If no, continue to step (f).

(i) A continuation sign is inserted between the previous phoneme and the current phoneme (step 29), and step (f) is continued.

For example, the phonetic symbol of "abdominal" is [ae-b-da-meh-nehl] before continuous sound processing and [ae-b-dam ~-~ mehn ~ -nehl] after continuous sound processing. do. The pronunciation symbol "farmerh" of "farmer" is [far-merh] before continuous sound processing, and becomes [farm---merh] after processing. The pronunciation symbol "gahvernmehnt" of "government" is [gah-verhn-mehn-t] before the treatment and [gah-verhn--mehn-t] after the treatment. The pronunciation symbol [eh'kohrdinglih] of "accordingly" is [eh-kohr-dihng-lih] before processing, and becomes [eh-kohr-dihng--lih] after processing.

Referring to FIG. 8, English is a multiple syllable language. Since each word can be broken down into several single syllables, it can be classified into thousands of basic voice units with different accents. Each speech unit has its own accent. Therefore, it is necessary to classify them appropriately.

The intonation classification step 3 classifies intonation of the phoneme string. The step (3) can be subdivided into an initial classification 31, a dual accentuation 33 and an advanced classification 34.

9, codes 1, 2, 3, 4, 03 and 04 are set for phonemes in a phoneme string. This includes the following steps.

(a) Check if the phoneme contains vowels (step 311). In that case, step d continues.

(b) The phoneme code is set to 1 (step 312).

(c) Select the next phoneme (step 313) and continue with step (a).

(d) Check whether the phoneme is the final phoneme containing the vowel (step 314). If no, continue to step k.

(e) Check that the phoneme contains accent marks (step 315). In such a case, the phoneme code is set to 3 (step 316), otherwise, the phoneme code is set to 4 (step 317).

(f) Check whether the phoneme is the final phoneme (step 318). In such a case, the process ends.

(g) Check if the next phoneme is unvoiced continuous sound (step 319). If no, continue to step (i).

(h) If the phoneme code is 3, change it to 03. If the phoneme code is 4, it is changed to 04 (step 320).

(i) Next phoneme is selected (step 321).

(j) Check if it is the final phoneme (step 322). If so, set the phoneme code to 4 and exit. If not, set the phoneme code to 1 (step 324) and continue with step (i).

(k) Check that the phoneme contains accent marks (step 325). In such a case, the phoneme code is set to 2 (step 326) and step (c) is continued. If no, set the phoneme code to 1 and continue with step (c).

Referring to FIG. 10, the double stress processing step 33 processes double stress in the phoneme string. This includes the following steps:

(a) Check if the phoneme code is 2 (step 331). If no, continue with step (f).

(b) Next phoneme is selected (step 332).

(c) Check if the phoneme code is 2 or 3 (step 333). If no, continue to step (g)

(d) Select the previous phoneme containing the vowel (step 334).

(e) Check if the phoneme code is 2 (step 335). In such a case, the process ends. If not, change the phoneme code to 2 (step 336) and continue with step (d).

(f) Check for the next phoneme (step 337). If so, the next phoneme is selected (step 338) and step (a) is continued; otherwise, the process ends.

(g) Check for the next phoneme (step 339). If yes, the next phone is selected (step 332) and step (c) is continued. If not, the process ends.

Referring to FIG. 11, the advanced classification is set to codes 5 and 6 for phonemes in the phoneme string. This includes the following steps:

(a) Check if the phoneme code is 2 (step 341). If no, continue to step (h).

(b) Select the next phoneme containing the vowel (step 342).

(c) Check if the phoneme code is 1 (step 343). If not, the process ends.

(d) Change the phoneme code to 6 (step 344).

(e) Select the next phoneme containing the vowel (step 345).

(f) Check whether the phoneme code is 1 (step 346). If not, the process ends.

(g) Change the phoneme code to 5 (step 347) and end.

(h) Check for the next phoneme (step 348). If yes, select the next phoneme and continue with step (a). If not, the process ends.

For example, the pronunciation symbol of "abandoned" is [eh'baendehnd], and after processing, it becomes [eh1-baen2-dehb4-d4]. The phonetic symbol of "actual" is ['aektshuehl] and becomes [ae2-k1-tshuehl4] after processing. The phonetic symbol of "appraisal" is [eh'prezehl] and becomes [eh1-pre2-zehl4] after processing. The phonetic symbol of "exquisite" is ['eakskwihziht] and becomes [ea1-k1-s1-kwih6-zih04-t4] after processing. The phonetic symbol of "absolutely" is ['aebsehlutlih] and becomes [ae2-b1-sehl∼6--lu5-t1-lih4] after the treatment.

The language synthesis step 4 uses phonemes as search units. Phonetic waveform data is retrieved from the decomposed phoneme sequence. The phonetic speech waveform database is extracted from the speech waveform originally recorded for a single word and indexed according to the phonemes.

The specification describes a method for decomposing and synthesizing English phonetic symbols. It decomposes known phonetic symbols into basic phoneme units of several parts. Speech waveform data of phonemes corresponding to the decomposed phoneme sequences are synthesized by language. This has the advantage of reducing the language data storage space and maintaining the accent of the original language.

Any modification that is considered to be within the spirit and scope of the appended claims will be apparent to those skilled in the art.

Claims (18)

In the method of decomposing and synthesizing English phonetic symbols, Phoneme decomposition step of converting English words to phoneme series, A continuous sound processing step of inserting a continuous sound at a continuous position in the phoneme sequence; Intonation classification step of classifying intonation of the phoneme string, and A language synthesis step of synthesizing phoneme voice waveform data of a phoneme corresponding to the phoneme string into a speech; Way. The method of claim 1, The phoneme decomposition step A preprocessing step of converting the phonetic symbols into phonetic symbol codes, combining / inserting / replacing the phonetic symbol codes, and calculating the number of consonants; A prefix consonant processing step of inserting diaeresis, if the phoneme separation location is located before the prefix consonant, If the phoneme separation location is located before the consonant between the vowels, the consonant processing step before the vowel inserting the diacritic marks, If the suffix consonant does not belong to the set consisting of [m], [n], [l], [r] and [ng], precede the suffix consonant with a syllable separating symbol and intonation code. Inserting a suffix consonant processing step Way. The method of claim 2, The pretreatment step Converting the phonetic symbol into the phonetic symbol code; Double consonants [tsh], [dsg], [pr], [br], [tr], [dr], [kr], [gr], [pl], [bl], [kl], [gl] , double consonant processing step that considers [tw], [hw], [sw], [dw], [kw], and [gw] etc. as a single inseparable consonant, A suffix consonant processing step of processing the suffix consonants [ts] and [dz] as a single non-separable consonant, a continuous consonant processing step formed by consonants other than [r] followed by consonants selected from the set consisting of [m], [n] and [l], If [s] is not a suffix consonant, then the unvoiced consonant processing step replaces the unvoiced consonants [p], [t] and [k] with voiced consonants [b], [d] and [g] immediately after [s]. , And Calculating the number of consonants, The continuous consonant processing step may include inserting [eh] between the continuous consonants so that the [m], when the continuous consonants are selected from the set consisting of end of word, after first accent, after second accent or after consonant; a consonant selected from the set consisting of [n] and [l] and the vowel [eh] followed by a non-r consonant Way. The method of claim 3, The consonant count step Counting prefix consonants, Counting consonants between vowels, and Counting suffix consonants Way. The method of claim 2, And wherein said prefix consonant processing step defines said phoneme separation location between a second prefix consonant and a final prefix consonant. The method of claim 2, The prefix consonant processing step (a) checking if the number of prefix consonants is at least 1, if not, stopping; (b) checking if the prefix consonant is the first prefix consonant, if not, continuing to step (d), (c) selecting the next prefix consonant, (d) inserting diacritics in front of the prefix consonant, and (e) checking that the prefix consonant is the final prefix consonant, and if so, stopping and continuing to step (c) if not Way. The method of claim 2, The inter-vowel consonant processing step places the phoneme separation position in front of the final vowel consonant and before the vowel consonant not belonging to the set consisting of [m], [n], [l], [r] and [ng]. Way. The method of claim 2, Consonant processing step between the vowels (a) checking the number of consonants between the vowels and continuing to step (e) if the number is 1 and stopping if the number is 0, (b) checking whether the consonant between the vowels belongs to a set consisting of [m], [n], [l], [r] and [ng] and inserting a syllable separator before the consonant between the vowels, (c) searching for consonants between the following vowels, (d) checking for consonant between the final vowels, if not continuing to step (b), and (e) inserting a syllable separator before the consonant Way. The method of claim 2, And the suffix consonant processing step determines the phoneme separation position before the final suffix consonant and before the suffix consonant that does not belong to the set consisting of [m], [n], [l], [r] and [ng]. The method of claim 2, The suffix consonant processing step (a) checking the number of suffix consonants and stopping if the number is zero, (b) checking whether the suffix consonants belong to a set consisting of [m], [n], [l], [r] and [ng], and if not, inserting syllable delimiters before the suffix consonants, (c) searching for the next suffix consonant, and (d) searching if the consonant is the final suffix, in which case it stops and if not, continuing to step (b); Way. The method of claim 1, And the continuous sound processing step is performed on consonants that are preceding consonants of the next phoneme while continuing to the previous consonants. The method of claim 11, The continuous sound processing step (a) checking the number of phonemes and stopping if the number is less than two, (b) selecting a second phoneme, (c) checking whether the first symbol belongs to the set consisting of [m], [n], [l], [r] and [ng], if not, continuing to step (f), (d) checking whether the last symbol of the previous phoneme belongs to the vowel, the set consisting of [l] and [r], if not, continuing to step (h), (e) inserting a continuation note and a consonant consonant between the previous consonant and the current consonant, (f) selecting the next phoneme, (g) checking if the selected phoneme is the final phoneme, if not stopping and if not continuing to step (c), (h) checking whether the final phonetic symbol of the previous phoneme belongs to the set consisting of [m], [n] and [ng], if not, continuing to step (f), and (i) inserting a consecutive phonetic symbol between the previous phoneme and the current phoneme and continuing to step (f) Way. The method of claim 1, The intonation classification step An initial categorization step of setting a code to 1, 2, 3, 4, 03, and 04 for the phonemes in the phoneme string; A dual stress treatment step of processing double stress in the phoneme string, and An advanced classification step of setting codes to 5 and 6 for phonemes in the phoneme string; Way. The method of claim 13, The initial classification step (a) checking that the phoneme contains a vowel, in which case continuing to step (d), (b) setting the phoneme code to 1, (c) selecting the next phoneme, continuing to step (a), (d) checking that the phoneme is the last phoneme containing the vowel, if not, continuing to step (k), (e) checking that the phoneme includes accented symbols, if so setting the phoneme code to 3, otherwise setting the phoneme code to 4, (f) checking that the phoneme is the final phoneme, and if so, stopping; (g) checking if the next phone is an unvoiced phone, if not, continuing to step (i), (h) changing to 03 when the phoneme code is 3 and changing to 04 when the phoneme code is 4, (i) selecting the next phoneme, (j) checking if the selected phoneme is the final phoneme, if so set the phoneme code to 4, otherwise set the phoneme code to 1 and continue to step (i), and (k) check that the phoneme contains accent marks, in which case set the phoneme code to 2 and continue to step (c), otherwise set the phoneme code to 1 and continue to step (c) Containing steps Way. The method of claim 13, The double stress treatment step (a) checking if the phoneme code is 2, if not, continuing to step (f), (b) selecting the next phoneme, (c) checking whether the phoneme code is 2 or 3, and if none, continuing to step (g), (d) selecting previous phonemes containing vowels, (e) checking that the phoneme code is 2, stopping in that case and changing the phoneme code to 2 and continuing to step (d), (f) checking for the next phoneme, if present, selecting the next phoneme and continuing to step (a), otherwise stopping; (g) checking for the next phoneme, if present, selecting the next phoneme and continuing to step (c), otherwise stopping Way. The method of claim 13, The advanced sorting step (a) checking that the phoneme code is 2, if not, continuing to step (h), (b) selecting the next phoneme containing the vowel, (c) checking that the phoneme code is 1, if not, stopping; (d) changing the phoneme code to six, (e) selecting the next phoneme containing the vowel, (f) checking if the phoneme code is 1, if not, stopping; (g) changing the phoneme code to 5 and stopping, and (h) checking for the next phoneme, if present, selecting the next phoneme and continuing to step (a), otherwise stopping Way. The method of claim 1, And the speech synthesis step decomposing the phoneme sequence based on phoneme as a basic unit and synthesizing speech waveform data of a corresponding phoneme into a language. The method of claim 17, The speech waveform data of the phoneme is extracted from the speech waveform data originally recorded for a single word.